What the work is
Help a leading AI lab teach its models to write real, machine-checked mathematics in Lean.
Overview
A leading AI lab is looking for Lean engineers, formal mathematicians and proof engineers to help its AI models state and prove mathematics correctly. You'll write and review Lean 4 proofs, turn informal math into precise formal statements, and help the lab's researchers judge whether a model's proof is not just accepted by the checker but actually proves the right thing. If you enjoy writing Lean, know your way around mathlib, and can explain why a formalization is faithful or subtly wrong, this role is for you. This is a part-time commitment of at least 20 hours per week, with the option to increase to up to 40 hours per week.
Key Responsibilities
- Write correct, idiomatic Lean 4 statements and proofs that compile against current mathlib, across areas such as algebra, analysis, number theory, combinatorics and logic.
- Formalize natural-language mathematics, from competition problems and textbook results to research-level lemmas, paying close attention to whether the formal statement matches the original.
- Review AI-generated Lean statements and proofs, find where they fail or prove the wrong thing, and give clear, specific written feedback.
- Help define guidelines and rubrics for proof quality, statement fidelity and mathlib conventions.
- Collaborate with other Lean engineers and the lab's researchers to keep standards consistent and keep raising the quality bar.
Who gets hired
Core Qualifications
- Hands-on experience writing formal proofs in Lean 4, for example mathlib contributions, a formalization project, a Lean library or tool, or autoformalization work.
- Comfort with mathlib and Lean 4 tactics, and with finding and using the right lemmas.
- A strong background in proof-based mathematics, theoretical computer science or logic, through a degree or a research record.
- The ability to turn a written statement and proof into a correct formal statement and a proof that checks.
- Ability to engage reliably for at least 20 hours/week during weekdays.
- Clear written communication and the ability to explain proof strategy and formalization choices precisely.
Nice to have: experience with other proof assistants or dependently typed languages (Coq/Rocq, Isabelle, Agda, Haskell), Lean metaprogramming, or AI-for-math work such as LLM provers, Lean agent environments, or benchmarks like miniF2F, ProofNet or PutnamBench. You don't need all of these to apply.
Pay
$90–110/hr, fully remote.